Modelling heterogeneity in scale directly: implications for estimates of influence in freight decision-making groups

The state of practice in the modelling of heterogeneous preferences does not separate the effects of scale from estimated mean and standard deviation preference measures. This restriction could lead to divergent behavioural implications relative to a flexible modelling structure that accounts for scale effects independently of estimated distributions of preference measures. The generalised multinomial logit (GMNL) model is such an econometric tool, enabling the analyst to identify the role that scale plays in impacting estimated sample mean and standard deviation preference measures, including confirming whether the appropriate model form approaches standard cases such as mixed logit. The GMNL model is applied in this paper to compare the behavioural implications of the minimum information group inference (MIGI) model within a study of interdependent road freight stakeholders in Sydney, Australia. MIGI estimates within GMNL models are compared with extant mixed logit measures (see Hensher and Puckett, 2008) to confirm whether the implications of the restrictive (with respect to scale) mixed logit model are consistent to those from the more flexible GMNL model. The results confirm the overall implication that transporters appear to hold relative power over supply chain responses to variable road-user charges. However, the GMNL model identifies a broader range of potential group decision-making outcomes and a restricted set of attributes over which heterogeneity in group influence is found than the mixed logit model. Hence, this analysis offers evidence that failing to account for scale heterogeneity may result in inaccurate representations of the bargaining set, and the nature of preference heterogeneity, in general.